Gemma 4 E4B vs MiniCPM5-2B
Wins 0 of 6 areas
—
Wins 6 of 6 areas
Coding · Agents · Reasoning · Facts · Long documents · Following instructions
MiniCPM5-2B is the stronger all-rounder.
Scores updated · 32 tests both models report · How we compare
Where each one wins
Tests won in each of the six areas where both have results. Each piece is one test, so a longer bar means more evidence; grey means the two scored within a point of each other.
- AgentsCarrying out multi-step tasks on its own03MiniCPM5-2B3 of 4 tests · 1 tie
- CodingWriting and fixing software03MiniCPM5-2B3 of 3 tests
- ReasoningHard problems that need careful thinking02MiniCPM5-2B2 of 3 tests · 1 tie
- FactsGetting facts right instead of making them up01MiniCPM5-2B1 of 2 tests · 1 tie
- Long documentsFinding answers in very long texts01MiniCPM5-2B1 of 1 test
- Following instructionsDoing exactly what it is asked01MiniCPM5-2B1 of 1 test
Long documents and following instructions rest on a single test each.
What it costs
Prices per million tokens, roughly 750,000 words. The bars show the cost of a million tokens read plus a million written.
The biggest differences
The tests each model wins by the widest margin, up to three each. Scores are out of 100.
Where MiniCPM5-2B pulls ahead
- Reasons across sets of long documentsAA-LCR+27.3points ahead
- Multi-step tasks using many real software toolsMCP Atlas+20.2points ahead
- Recent programming contest problemsLiveCodeBench v6+18.5points ahead
Where Gemma 4 E4B pulls ahead
No clear win on a test scored out of 100.
Every test, side by side
All 32 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingMiniCPM5-2B
- LiveCodeBench v6MiniCPM5-2B by 18.55270.5+18.5
- Terminal-Bench 2.1MiniCPM5-2B by 6.71.98.6+6.7
- SciCodeMiniCPM5-2B by 1.924.426.3+1.9
AgentsMiniCPM5-2B
- MCP AtlasMiniCPM5-2B by 20.26.226.4+20.2
- τ-Bench V3 · BankingMiniCPM5-2B by 15.45.420.8+15.4
- GDPValMiniCPM5-2B by 10.4010.4+10.4
- Terminal-Bench 4.0tie00tie
ReasoningMiniCPM5-2B
- GPQA DiamondMiniCPM5-2B by 12.657.670.2+12.6
- Humanity's Last ExamMiniCPM5-2B by 5.13.88.9+5.1
- CritPttie0.60.3tie
FactsMiniCPM5-2B
- AA-Omniscience · Non-hallucinationMiniCPM5-2B by 969.178.1+9
- AA-Omniscience · Accuracytie8.68.4tie
Following instructionsMiniCPM5-2B
- IFBenchMiniCPM5-2B by 17.944.262.1+17.9
Other results18 tests, not counted
Tests outside the eight areas. They are not counted above: several are summary scores built from other tests, or the same test under another name.
- τ³-bench (4 domains), averageMiniCPM5-2B by 35.933.569.4+35.9
- RULER 128KMiniCPM5-2B by 19.218.838+19.2
- BFCLv4MiniCPM5-2B by 16.346.462.6+16.3
- SWE-bench Multilingual (OpenCode)MiniCPM5-2B by 150.715.7+15
- Claw-Eval General (EN), pass^3MiniCPM5-2B by 12.222.434.6+12.2
- AA-OmniscienceMiniCPM5-2B by 8.1-19.7-11.7+8.1
- WebArena-VerifiedGemma 4 E4B by 817.19.1+8
- AA Agentic IndexMiniCPM5-2B by 7.21.89+7.2
- Terminal-Bench 2.1 (OpenCode)MiniCPM5-2B by 5.605.6+5.6
- HLE (with tools)MiniCPM5-2B by 5.5611.5+5.5
- Artificial Analysis Coding IndexMiniCPM5-2B by 5.19.414.5+5.1
- Claw-Eval General (EN), averageMiniCPM5-2B by 4.656.160.7+4.6
- BrowseComp-Plus (OpenClaw subset)Gemma 4 E4B by 4.22116.8+4.2
- AA IntelligenceMiniCPM5-2B by 3.68.912.5+3.6
- EnterpriseOps-GymMiniCPM5-2B by 3.59.913.4+3.5
- AutomationBench PublicMiniCPM5-2B by 2.81.94.7+2.8
- LiveCodeBench Pro 25Q2tie29.929.3tie
- IFEvaltie87.987.8tie
Questions people ask
Which is better, Gemma 4 E4B or MiniCPM5-2B?
MiniCPM5-2B wins all six areas where both have results: coding, agents, reasoning, facts, long documents and following instructions. Gemma 4 E4B wins none.
Which is better for coding?
MiniCPM5-2B. It wins 3 of the 3 coding tests both models report; Gemma 4 E4B wins none.
How do you compare the two?
We use the 32 benchmark tests both models have published scores on. The verdict counts the 14 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 18 are listed but not counted, because several are summary scores or repeat a test. Each score is the one shown on the model's own page.